1. Field of the Invention
The present invention relates to a noise reduction device and a noise reduction method of compression coded video signals.
2. Related Art
In a digital television broadcast, video data of large quantity and high quality can be broadcast in limited transmission capacity by transmitting compressed stream data of video signals which is compression coded in MPEG coding method. Also when recording video signals by DVD recorder or the like, the video signals are once compressed in MPEG coding method and then recorded, so that video signals of high quality may be recorded for a long time (see non-patent document 1). However, for transmitting or recording a larger quantity of video data, it is required to compress and code data at higher compression rate, and thus occurrence of coding noise due to compression and coding cannot be avoided.
Block noise which appears to be block stripes is known well as MPEG coding noise. Further, pulsing noise occurs, which provides rough-textured image caused by temporal difference in coding noise occurring in each image.
Noise reduction methods of MPEG coded image are roughly classified into two-dimensional noise reduction process (2DNR) for reducing plane noise occurring in the image, and three-dimensional noise reduction process (3DNR) for reducing temporal noise occurring between the images.
Specifically, 2DNR is effective for reducing the block noise, and 3DNR is effective for reducing the pulsing noise, and 2DNR and 3DNR are combined to be used for reducing noise.
Noise is basically reduced by a low pass filter for removing high frequency components. However, if uniformly processed by a low pass filter, blurry image may occur in 2DNR, or retained image noise trailing the motion may occur in 3DNR, and these are adaptively processed by filtering.
As an example of adaptive process, it has been proposed to detect the motion amount of the image, and intensify the 3DNR and weaken the 2DNR in a still image which has small motion amount (see, for example, patent document 1).
Further, it has been proposed to reduce the noise while suppressing the occurrence of retained image noise also in an image fast in motion, by detecting the motion amount of the image and the correlation amount between frames, and controlling the 2DNR and 3DNR according to the motion amount and the correlation amount between frames (see, for example, patent document 2).
It has also been proposed to control the intensity of noise reduction filter to be executed before the coding process according on the motion vector detected upon compression coding of the image (see, for example, patent document 3).
In another proposal, synthesizing ratio of the image processed by 2DNR and the image processed by 3DNR is controlled depending on the motion of the image, and in the image of fast motion, the effect of 3DNR is suppressed, while the effect of 2DNR is emphasized (see, for example, patent document 4).
Thus, in the case of 3DNR, it is a general way to detect the motion of the image, intensify the effect of 3DNR when the motion is small, and weaken the effect when the motion is large so that the temporal noise is reduced and the retained image noise is suppressed.
Patent document 1: JP-A-11-69202
Patent document 2: JP-A-2003-348383
Patent document 3: JP-A-2003-333370
Patent document 4: JP-A-2005-150903
Non-patent document 1: ISO/IEC 13818-2 (MPEG2 Video)
The motion is detected by the amount of temporal change of pixel level. As the target image has more high frequency components (sharper image), the level change amount in the image is larger, and hence the temporal level change amount of pixel also increases, so that the motion is more likely to be detected. That is, the image having more high frequency components provides substantially higher motion detection sensitivity. To the contrary, the image having fewer high frequency components (blurry image) has small level change amount in the image, and hence the temporal level change amount is also smaller, so that the motion is less likely to be detected. That is, the image having fewer high frequency components provides substantially lower motion detection sensitivity.
Since 2DNR includes a low pass filter as a basic component, the high frequency components are lost when 2DNR is applied in order to reduce coding noise such as block noise occurring more. Since 3DNR has high motion detection sensitivity to a sharp image having many high frequency components, the adaptive process is done effectively, and the noise can be reduced without causing retained image noise. However, in the image of which high frequency components are lost due to strongly applying of 2DNR, the motion detection sensitivity becomes lower, and thus 3DNR is applied strongly so that the retained image noise is more likely to occur.
Accordingly, when 2DNR is emphasized, the coding noise is reduced more securely, but retained image noise is caused by the subsequent 3DNR process.